Tracking health outcomes in longitudinal biomedical studies has important clinical implications in preventive medicine, which is methodologically challenging because of time-varying covariates and unbalanced longitudinal data. A tracking index, the ``tracking prediction probability" (TPP), is proposed to quantify the ability of a subject developing certain health outcomes conditioning on his\her health outcomes and covariates at a past time point. For the unknown unstructured distribution functions in practice, we consider a class of structured dynamic copula-based models for the TPPs at two time points, and give the continuous-time TPP estimates of TPP through local smoothing allowing for the known or unknown copula family. They can provide additional insights into the dynamic patterns of the disease risk factors and the health outcomes. We develop the asymptotic properties of the discrete-time and continuous-time estimators for the copula parameters and the TPPs, and investigate the finite sample properties of these estimators through a comprehensive simulation. Applying our models and estimation procedure to a large epidemiological study of child growth and cardiovascular health, the dynamic relationships between the risk of developing hypertension in the future and the evolving patterns of past blood pressure levels and BMI changes are efficiently illustrated.